Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down at the end of May, and its replacement, statalist.org is already up and running.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: Panel Regression with missing x for one entity


From   Nick Cox <njcoxstata@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: Panel Regression with missing x for one entity
Date   Thu, 16 Aug 2012 15:42:01 +0100

My guess is that many of us remain puzzled about precisely what you
want to know and also have difficulty in assessing what is obvious to
you, as the thread already appears to have cycled around the same
points about twice. My take is that

1. The model as fitted used no observations (cases, records) in which
values were missing on any of the variables named. That's an absolute
for Stata.

2. That does not stop the model being used for out-of-sample
predictions, but you would need to supply values somehow for the
predictors that were missing in your data.

3. Zero does not qualify under #2 unless you have substantive grounds
for supposing that missing really means zero, in which case you could
go back to refitting the model with missings filled in as zero.

4. If you don't know what the missings should be, I don't know what
you should do, except perhaps look at countries which on other grounds
you expect to behave similarly. That's a matter of economic judgment
for you or your colleagues.

Nick

On Thu, Aug 16, 2012 at 3:23 PM, Whelan, Paul
<Paul.Whelan@bankofengland.co.uk> wrote:

> Does anyone have some advice re: the below? All countries were included in the estimation,
> just that at the prediction stage in there happens to be missing data on the right hand
> side then stata gives me no forecast?

Whelan, Paul

> The is no information for this particular conditioning variable contributing
> to the forecast for this country but it did contribute to the fit for the other
> countries. So, a forecaster should still be able to make a projection. It's like
>
> y_1 = c_1 + beta_11 X_1 + beta_12 X_2 + eps_1
> y_2 = c_2 + beta_12 X_1 + + eps_2
>
>
> stata doesn't compute E[y_2(t+h) | X_1] if [c_2 , beta_12 ] is estimated from the panel

Nick Cox

 It's often difficult to know what is obvious, but let's spell out that
> if any values were missing for a particular country, then no data for
> that country contributed to the model fit.

Whelan, Paul

>> Sure - I guess that is obvious.

Maarten Buis

>> On Tue, Aug 14, 2012 at 7:43 PM, Whelan, Paul wrote:
>>>> I'm running a panel regression with a cross-section of countries:
>>>
>>> y_it = c_it + beta_1 x_1 + ... + beta_n x_n + eps_it
>>>
>>> For one of my countries , a particular x is completely missing.
>>> The estimation works fine but I cannot produce predictions for this particular entity.
>>
>> You (through Stata) create the prediction by just filling in the
>> equation you gave above. A missing value just mean you don't know the
>> value. What would be the outcome of: beta_1*"an unknown value"? Stata
>> correctly answers: "an unknown value".
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/


© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index